JahangirBabar.

Building the AI research infrastructure behind a media-studies lab — turning fragmented archives into something you can actually ask questions of.

• Built a mobile-gaming research archive indexing 1,000+ titles with Next.js, TypeScript, and DynamoDB, making historically fragmented metadata searchable.
• Cut researcher lookup time ~70% by replacing keyword search with Gemini function-calling for cross-entity semantic queries across games, devices, and publications.
• Reduced interaction friction ~60% with a voice-driven conversational interface built on Whisper and ElevenLabs.

An AI discovery platform that watches the research landscape so a 40+ lab institute doesn't have to.

• Eliminated ~80% of manual research-monitoring work with an AI dashboard aggregating publications, events, and media mentions across 40+ robotics labs.
• Improved surfaced-signal quality 3× with a Tavily + OpenAI ranking pipeline that filters noisy web results.
• Built a Flask scanning engine with SQLite persistence and a React review interface for incremental scans and approvals.

Owned authentication & onboarding UX on a CIAM product serving millions — where a single friction point compounds across every login.

• Reduced failed authentication attempts ~35% across 5 enterprise services by mapping Okta login friction and redesigning error-handling prompts with engineering.
• Increased onboarding completion ~20% for 2M+ users by analyzing drop-off in Pendo and FullStory and prioritizing registration & auth UI changes.
• Shortened average login resolution time ~40% by mapping end-to-end auth journeys and driving changes validated through behavioral analytics.

Backend and ML-adjacent work on AWS microservices powering enterprise analytics and prediction workloads.

• Increased platform throughput 20% contributing to an AWS-based microservices architecture for analytics and prediction.
• Cut inference latency from ~500ms to ~50ms by introducing FastAPI alongside Django and optimizing batch-prediction schemas.
• Lifted user engagement 40% with an LLM-powered content-generation feature producing user-specific outputs in-app.
• Removed 15+ hours/week of manual data prep with Python automation and containerized execution pipelines.

Developer tooling and CI for a graph-database management interface — making a complex SQL/Cypher product easier to trust.

• Increased feature adoption 25% by rebuilding pgAdmin UI components in React and TypeScript for the AgensSQL graph DB interface.
• Cut regression-testing cycle time ~50% with GitHub Actions pipelines validating Cypher queries against PostgreSQL compatibility.
• Reached 95%+ test-environment reliability with Python scripts seeding and validating vertex-edge datasets via psycopg2.